On Tue, Aug 12, 2008 at 11:55 PM, Wes Hardaker <
[EMAIL PROTECTED]> wrote:
> Right, but it *does* still still find it in the non-standard path if the
> standard one doesn't exist which at least makes me think the search path
> actually contains the new extraction (somehow; again I'm clueless on the
> extraction steps). If the new paths are added in the front of the
> search path instead of the end, the problem would likely go away (and
> leaving the system paths at the end would likely still allow some of the
> bootstrapping issues to not reappear (assuming there are any; again: I
> don't know))
I just tried this on a Solaris box with totally different results. Latest
versions
of PAR, PAR::Packer, Module::ScanDeps and there shouldn't be any
significant differences between Solaris and Linux.
On Solaris the early extracted modules have there names mangled, e.g.
IO.so is extracted as 52934cf9.so. And I see that the executable is actually
looking for the mangled names, strace (truss) doesn't show any open(),
stat() or similar call for something named ".../IO.so". This is in sharp
contrast
to the strace on Linux that has
3260 stat64("/usr/lib/perl/5.10/auto/IO/IO.so", {st_mode=S_IFREG|0644,
st_size=15868, ...}) = 0
The only significant difference is the version of Perl itself. The Solaris
box
has 5.8.6, my Linux box has 5.10.0.
Maybe PAR does something that used to work, but not anymore in 5.10?
It definitely does mung Perl internals in the area of XS module loading:
- PAR::Packer patches the copy of XSLoader.pm that it packs into every
executable
- PAR::Heavy (from PAR) overrides functions from DynaLoader with its own
versions
Cheers, Roderich